Classificação de células brancas a partir de amostras de sangue usando rede neural artificial convolucional

Abstract

The complete blood count, a microscopic exam of human blood samples, is a test commonly used to asssess state of health, organism changes and diagnose diseases in pacients. The exam consists of analyzing the blood components, counting and classifying the elements between red part (red blood cells) and white part (white blood cells and platelets). The complete blood count is performed in medical laboratories by specialized professionals, demanding a lot of time and high cost equipment and maintenance. As an alternative, this project proposes a Deep Learning computational model, capable of classifying white cells in blood sample images automatically. Utilizing Convolutional Neural Network tools, the proposed model achieved 97% accuracy in white cell classification, using a labeled public testbase. However, when applying the model to images outside the trained pattern, the accuracy was not maintained. A web prototype was developed to the present the classified data and where the professional can select an image to assert the recognized data for decision making in the elaboration of a complete blood count.